Cheaper, Quicker, Almost as Good

Hello,

This is Simon with the latest edition of The Weekly. In these updates, I share key AI related stories from this week's news, list upcoming events, and share any longer form articles posted on the website.

P.S. — sharing this issue with a couple of people gets you a free guide. More on that partway down.

Across nearly 100 editions of Plain AI, I've never focused on the latest AI models coming to market. The changes are so rapid that it's very hard to stay on top of them, and I'm not overly interested in how a model scored on a certain benchmark test. I've always been more concerned with how a model's output worked for me on a specific task. That's the real-world test that I, and most readers like you, care about.

That said, it has been interesting to see Anthropic release their latest model, Opus 5.5. The detail that caught my eye was that Anthropic describes it as almost comparable to their most powerful model (Mythos 5.1), while being quicker and up to 40% cheaper than Opus 5 on typical workloads. There's certainly a race between the major players to become the dominant choice, which helps keep prices and performance competitive, but building a model that's close to your very best at a lower cost is notable. It makes me wonder whether we're starting to see a new phase, where the cost of running these models begins to come down.

There's been plenty of press coverage in recent months about spiralling AI costs as more people and companies rely on AI for more aspects of their work. For senior managers in Finance teams, cheaper and quicker models must sound like music to their ears.

It also reminds me of the early days of mobile phones, when we paid quite a lot for texts and calls. They were premium items back then, but over time the economics improved, and today the vast majority of mobile phone contracts don't charge for them at all. We're clearly a long way from that with AI, but at some point I'm sure a basic level of AI tools will become a commodity that doesn't warrant a price premium.

Every customer signal is captured in one shared context layer, always current and compounding. Agents and workflows build pipeline, chase every buying signal, and move deals forward, an always-on revenue engine running alongside your team.

With Attio, you’ll get:

Teams like Parallel, Turbopuffer, and Wordsmith build on Attio. Are you one of them?

Get started now

Workday published its annual Global Workforce Report on Monday, drawing on data from over 550 employers and surveys of 6,000 full-time workers. It showed the demand for AI engineering skills rose 51% between September 2025 and July 2026. Demand for basic AI prompting skills fell 25% over the same period, suggesting that entry-level AI literacy is already being absorbed into software itself.

The reskilling gap sits just beneath that data: 79% of workers said they know which skills they need to develop, but only 66% said their employer is actively helping them do so. Internal promotions were flat year-on-year. Four in ten employees went through a reorganisation or restructuring in the past twelve months.

Why it matters: Organisations have spent two years teaching staff to write prompts. This report suggests that phase is largely over, and the next one, which includes building, deploying, and integrating AI into processes, requires a different level of investment.

Microsoft had been on track to spend approximately $1 billion this year on internal use of Anthropic's technology, but has since cut that projection by more than a third, according to The Information, reported on 6 October. Meta went further: the number of employees using Claude Code fell from around 60,000 to roughly 30,000 after Meta began steering staff toward its own internal tools, Muse Code and MetaCode. Microsoft is pushing engineers toward GitHub Copilot.

The timing is notable: the moves come as Anthropic heads toward an IPO, and as a Wall Street Journal report found that only 11% of nearly 400 surveyed businesses could accurately forecast their own AI tool spending. Meta had been on track to spend billions on internal AI tools, with Claude Code its most used.

Why it matters: When companies the size of Microsoft and Meta find that third-party AI costs are hard to predict and control at scale, the natural response is to redirect spend to tools they own. It is an early signal of a pattern that will become more common: enterprises running third-party models for specific use cases while building proprietary stacks for heavy internal usage. For anyone selling AI tools into large organisations, cost predictability and stack control are now as important as raw capability.

Venture-backed AI companies acquired 195 AI startups through 29 September 2026, already exceeding the total for all of 2025 by 14%, according to a Crunchbase review published this week. The number of companies doing the buying rose only 2%, suggesting a small group of well-capitalised players is driving the majority of activity rather than a broad market of acquirers. The pace has picked up sharply as early-stage valuations compress and the market for standalone AI point solutions becomes harder to sustain independently.

Why it matters: Rapid consolidation changes the vendor landscape in ways that matter to every organisation currently evaluating or buying AI tools. Products acquired today may be absorbed, rebranded, or discontinued within twelve months. For IT, procurement, and operations teams, the long-term viability of a vendor is now a reasonable, practical due-diligence question. The market is sorting itself into a smaller number of platforms

Thanks for reading, and see you next Thursday.

Simon,

Was this email shared with you? If so subscribe here to get your own edition every Thursday.

You can also follow me on social media: